Slow Productivity AI

to free us from meaningless tasks

Piotr Stepinski

Infinitii ai

2025-04-25

About Me

  • CTO at infinitii ai
  • GitHub: stepinski
  • LinkedIn: stepinsky
  • Email: stepinski@gmail.com

Agenda

  1. The Slow Productivity Philosophy
  2. My “Why”: Personal Productivity Challenges
  3. My Core Belief: AI-Enhanced Deep Work
  4. The Technical Vision & Implementation
  5. Solution Demonstration: Key AI Pipelines
  6. Conclusion & Resources

The Slow Productivity Philosophy

Cal Newport’s Concept

“In his book Slow Productivity, Cal Newport argues that modern work culture prioritizes busyness over effectiveness, leading to stress, shallow work, and burnout.”

  • Natural pace
  • Meaningful work focus
  • Distraction reduction

My “Why”

The Problem with Personal Productivity

Daily routine struggles:

  • Writing everything in notes but lacking a consistent system
  • Trying various tools but always reverting to basic note-taking
  • Scattered information across notes, emails, and meeting records

System failures:

  • Starting productivity systems but failing to maintain them
  • Difficulty finding and connecting related information

The Note-Taking Struggle

# What my notes look like:
notes = [
    {"date": "2025-03-15", "content": "Team mtg - follow up API keys"},
]

The Note-Taking Struggle

# What my notes look like:
notes = [
    {"date": "2025-03-15", "content": "Team mtg - follow up API keys"},
    {"date": "2025-03-17", "content": "idea for deplymnt process"},
]

The Note-Taking Struggle

# What my notes look like:
notes = [
    {"date": "2025-03-15", "content": "Team mtg - follow up API keys"},
    {"date": "2025-03-17", "content": "idea for deplymnt process"},
    {"date": "2025-03-20", "content": "todo: revew PR for pipeline"}
]

# The problem:
# - Unstructured and disconnected
# - No time for proper review and organization
# - Valuable insights get lost

This isn’t just a personal problem. Studies show knowledge workers spend up to 20% of their time searching for information they’ve already seen or created.

My Core Belief

AI Can Free Our Creative Potential

Through AI we can automate the dull and time-consuming stuff to free up more creative potential of our minds

  • Notes organization shouldn’t require manual effort
  • Task extraction can be automated
  • Knowledge connections can be discovered automatically
  • All while maintaining privacy and control

How This Aligns with Slow Productivity

  • Automate the shallow work
    • Let AI handle organizing, connecting, task-tracking
  • Preserve mental space for deep work
    • Free from organizational overhead
  • Reduce cognitive load
    • One system instead of many disconnected tools

The Technical Vision

Ideal Workflow Components

Slow Productivity AI Architecture

Unified Knowledge Sources

Your Digital Life

Unified Knowledge Sources

Your Digital Life

  • Obsidian notes (primary knowledge base)
  • Supernote handwritten content (OCR processed)
  • GitHub Issues (work tasks)
  • Email and Teams conversations (transcribed)
  • Browser bookmarks and read-later content

The magic happens when these sources are connected through a single semantic layer, allowing cross-referencing and discovery across previously siloed information.

Core Processing Components

Component Role Slow Productivity Benefit
Hugging Face Transformers Local NLP for meaning extraction Privacy, no context-switching
n8n Self-hosted automation workflows Control, no third-party dependencies
Obsidian + GitHub Issues Knowledge and task management Centralized system, reduced tool switching
Local LLM Same engine for all tools Consistent context across interfaces

Implementation Details

The Local LLM Setup

# Core LLM setup
def setup_local_llm():
    # Choose model based on your needs
    pass

The Local LLM Setup

# Core LLM setup
def setup_local_llm():
    # Choose model based on your needs
    """
    Options:
    - Llama 3 8B: Good balance
    - Mistral 7B: Efficient
    - Phi-3 mini: Lightweight
    """
    pass

The Local LLM Setup

# Core LLM setup
def setup_local_llm():
    # Choose model based on your needs
    """
    Options:
    - Llama 3 8B: Good balance
    - Mistral 7B: Efficient
    - Phi-3 mini: Lightweight
    """
    
    # Install Ollama
    # curl -fsSL https://ollama.com/install.sh | sh
    
    # Start service
    # ollama pull llama3:8b
    # OLLAMA_HOST=localhost:11434 ollama serve

The choice of LLM model is crucial - it affects both performance and quality. For most MacBook Pro users, Llama 3 8B provides the optimal balance between speed and capability.

Notes Processing with LlamaIndex

from llama_index import VectorStoreIndex, SimpleDirectoryReader
from llama_index.llms import Ollama

def process_obsidian_notes():
    # Path to vault
    OBSIDIAN_PATH = "/path/to/obsidian/vault"
    
    # Load documents
    documents = SimpleDirectoryReader(
        OBSIDIAN_PATH, recursive=True
    ).load_data()
    
    # Connect to LLM
    llm = Ollama(model="llama3:8b", url="localhost:11434")
    
    # Create index
    index = VectorStoreIndex.from_documents(documents)
    
    # Extract tasks
    tasks = []
    for doc in documents:
        if "TODO" in doc.text or "- [ ]" in doc.text:
            tasks.extend(extract_tasks(doc.text, llm))
    
    return tasks

n8n Automation Workflow

n8n Workflow for Task Extraction

  • Scheduled workflows run during non-focus time
  • Event-based triggers for new notes or changes
  • Multi-step processing to extract and structure tasks

Workflows run while you sleep, so all your insights are processed and ready by morning.

Task Management Integration

# GitHub Issues integration
def create_github_tasks(tasks):
    from github import Github

Task Management Integration

# GitHub Issues integration
def create_github_tasks(tasks):
    from github import Github
    
    # Connect to GitHub
    g = Github(os.environ.get("GITHUB_TOKEN"))
    repo = g.get_repo("username/tasks-repo")

Task Management Integration

# GitHub Issues integration
def create_github_tasks(tasks):
    from github import Github
    
    # Connect to GitHub
    g = Github(os.environ.get("GITHUB_TOKEN"))
    repo = g.get_repo("username/tasks-repo")
    
    for task in tasks:
        # Avoid duplicates
        if not is_duplicate_task(repo, task["title"]):
            # Create issue
            repo.create_issue(
                title=task["title"],
                body=f"Source: {task['source']}",
                labels=[task["priority"]]
            )

Solution Demonstration: Key Pipelines

Demonstration Approach

Demo Flow Visualization

Pipeline 1: Notes Improvement

Before:

Meeting with team (04/12)
discussed pipeline issues - john mentioned 
permissions problem

Pipeline 1: Notes Improvement

Before:

plan:
- Montreal -> new model (testing and fixing) -> goal by EOW to setup next meeting
- Tesler pipelines -> technical debt due to iandi new deployments and scaling
- IandI in Tesler - weekly dwp

Pipeline 1: Notes Improvement

Before:

plan:
- Montreal -> new model (testing and fixing) -> goal by EOW to setup next meeting
- Tesler pipelines -> technical debt due to iandi new deployments and scaling
- IandI in Tesler - weekly dwp

Project Plan Update

Metadata

Property Value
para_category 3-Resources
type note
status active
enhanced_date 2025-04-16 19:19:09
enhanced_method llm

Current Focus – Montreal

New model development, testing, and fixing
Goal: Have everything set up for the next meeting by EOW (End of Week)

  • Complete testing and fixing for the new model
  • Set up next meeting with all necessary materials ready

Tesler Pipelines – Technical Debt

Due to recent IandI deployments and scaling, there is a need to address technical debt in Tesler pipelines.

  • Review and prioritize technical debt items
  • Develop a plan to resolve the technical debt

IandI in Tesler – Weekly DWP

Continue to monitor and manage IandI’s Daily Work Plan (DWP) on a weekly basis.

Source

[[ocm today]]

The transformation happens automatically without you having to change your note-taking behavior. Write naturally, and let AI do the structuring.

Pipeline 2: Project Summary Report

n8n Workflow:

+------------------+
| Obsidian Vault   |
| Change Trigger   |
+--------+---------+
         |

Pipeline 2: Project Summary Report

n8n Workflow:

+------------------+
| Obsidian Vault   |
| Change Trigger   |
+--------+---------+
         |
         v
+--------+---------+
| Local LLM Node   |
| (Extract Tasks)  |
+--------+---------+
         |

Pipeline 2: Project Summary Report

n8n Workflow:

+------------------+      +----------------------+
| Obsidian Vault   |      | GitHub Issues API    |
| Change Trigger   |      | (Create Issue)       |
+--------+---------+      +-----------+----------+
         |                            |
         v                            |
+--------+---------+                  |
| Local LLM Node   |                  |
| (Extract Tasks)  |                  |
+--------+---------+                  |
         |                            |
         v                            |
+--------+---------+                  |
| Task Processing  |                  |
| (Format & Sort)  |                  |
+--------+---------+                  |
         |                            |
         v                            |
+--------+---------+  +---------------+----------+
| Filter Tasks     +->+ Create GitHub Issue      |
| (Priority > Med) |  | for Each Extracted Task  |
+------------------+  +--------------------------+
  • Scheduled workflow runs daily at 6am (before work)
  • Processes new notes and updates since last run
  • Generates executive summary of active projects

Pipeline 3: Daily Template

# Daily Focus (2025-04-25)

## Priority Tasks
1. [API-123] Finalize pipeline params

Pipeline 3: Daily Template

# Daily Focus (2025-04-25)

## Priority Tasks
1. [API-123] Finalize pipeline params
2. [INFRA-45] Review K8s configs 
3. [TEAM-12] Prepare sprint planning

## Updates
- PR #234 merged (ML pipeline)

Pipeline 3: Daily Template

# Daily Focus (2025-04-25)

## Priority Tasks
1. [API-123] Finalize pipeline params
2. [INFRA-45] Review K8s configs 
3. [TEAM-12] Prepare sprint planning

## Updates
- PR #234 merged (ML pipeline)
- QA checks passing
- Docs ready for review

## Focus Blocks
09:00-11:00: Deep work - API-123
13:00-14:30: Team meeting
15:00-17:00: Deep work - INFRA-45

Research shows that simply having a clear plan reduces decision fatigue by up to 40% and can save 30 minutes of planning time daily.

Conclusion

The Slow Productivity AI Advantage

  • Consistency without effort - maintain notes without organizational overhead
  • Privacy by design - fully local processing, your data stays yours
  • Reduced cognitive load - automation frees mental space for creative work
  • Deep work preserved - minimize context switching and distractions

“The goal isn’t faster productivity, but deeper, more meaningful productivity.”

Key Takeaways

  1. Slow productivity is about meaningful output, not speed
    • AI should reduce friction, not add more digital noise
    • Automation should serve deep work, not interrupt it
  2. Privacy and ownership matter
    • Local, open-source LLMs make this possible today
    • Your knowledge base remains fully under your control
  3. Start small, but think holistically
    • Begin with one pipeline that solves your biggest pain point
    • Build toward a system that works the way your mind works

You don’t need to overhaul your entire workflow at once. Start with one pipeline, prove its value, and gradually expand your system.

Resources